Z-Image is here, and it's pretty damn great. Let's make it more NSFW!
Recommended Settings
Steps: 12
CFG: 1.0
Sampler: dpmpp_sde
Scheduler: simple
FP8 Download
For some reason, when you upload multiple formats (BF16 + FP8) on one model, which they now support, they hide the FP8 under the BF16 model group in the side menu...
ComfyUI
Models should be placed in diffusion_models, and uses the official workflow with the Load Diffusion Model loader, and loading the VAE/CLIP models separately. Models named "_AIO" should go in the checkpoints folder, and uses the normal checkpoint loaders.
Description
FAQ
Comments (29)
Can this be used with a character lora yet?
As long as you use a correct lora loader that handles z-image appropriately, yes.
Alternatively, you can mutate the layers yourself and re-save the character LoRA.
https://github.com/shootthesound/comfyUI-Realtime-Lora
Inspire Pack's block weight lora loader seem not to work with Z LoRas
@6tZ what do you recommend for training a compatible character Lora? I've been using mostly default settings in ai toolkit with mixed (bad) results. They work well in vanilla z-image, but not this one unfortunately.
@uncropped Hmm, I'll see if I can fix it on the checkpoint side for future versions. I have an idea.
Regarding how to train compatible, I don't know, normal settings in ai-toolkit should work. Didn't try myself, but I've tested on one and it worked with proper z-image LoRA loading (where it mutates the layers).
@6tZ I've found character Lora works pretty well if you use Qwen3 4b Abliterated as a text encoder.
@uncropped Huh?! Interesting, how the hell does this work any differently?
its already at 480,000 plz release it plz
No! It needs 310k more buzz
I've been testing some text encoders. Any chance to release 2.1 diffusers only?
Yeah 2nd, this. For the sake of two extra nodes AIO is hardly that much more convenient.
@hellyeah354 Sure, I will convert and upload!
There's an alternative: Use only the "model" output of the checkpoint loader (the CLIPs and VAE won't be loaded). You can continue using your custom CLIPs/VAE as usual. The only drawback is the unnecessary download time/disk space linked to the AIO checkpoint.
Sorry for the delays, busy working on new versions. I'll try to get 2.1 and 2.2 uploaded tomorrow.
is there any system prompt to generate prompts for your model?
Huh? What do you mean? No. This is an image model. You have to create system prompts for language models yourself :)
what vae and clip is included in the AIO? 3_4 and ae?
(hijacking the response to ask what's 3_4 ? I'm new to PONY and havent touched SDXL ecosystem in years, last time i delved in it there was but one VAE and a VAE fp16fix...)
EDIT : nevermind the SDXL content, it's late and i'm doing too much at a time lol
@FidgetyPenguin the text encoder is named qwen_3_4b. i didn't remember the full name but remembers the 3_4 numbers. I read that different text encoders and/or vae can affect the ability of the nsfw ability. I wanted to see which ones were used. If they just used the ones I mentioned then no point downloading the AIO.
@Melodic_Possible_582589 Ah right thanks for your answer (I though I was on a SDXL model lmao). you should be looking on Hugginface for the Qwen3 Abliterated or Heretic versions, they're pretty handy when it comes to NSFW. I think this is the one i use, (not sure if linking is allowed) svjack/Qwen3-VL-4B-Instruct-heretic-7refusal · Hugging Face
@Melodic_Possible_582589 It's the original models, qwen34b and flux vae. I just saved it as a checkpoint in comfy.
@FidgetyPenguin How to you use the linked model as VAE?
The checkpoint we've ALL been waiting for 😁 Alcaitiff is definitely a legend too. I currently using the great Jib mix ZIT too. It's very good.
Need to try that one. Jib's stuff is usually great!
And Alcaitiff is the GOAT!
these preview photos look absolutely phenomenal
Thank you kindly
The model doesn't always output this good of course, it's cherry-picking. But it's capable as hell!
@6tZ cherry-picking or not, as you said the capability is there! great job :)
Did Aliens create Z image Turbo and are they currently cooking up the base model with NoobAI, Pony, Illustrious, SDXL and Danbooru data set especially NSFW stuff or the whole of CivitAI. Who knows but my friend Steve say's he's a alien and he said he knows someone who works for Alibaba. The alienbaba man knows a secret agent from Mars who works for Tongyi-MAI. 😁 They say we'll done too. 👍
is it for base? I thought the base was just a 100 step standard of turbo, but they are linking up with the Noob/Illu guys for some other stuff